An interpretable nuclear power plant valve fault diagnosis method, system and medium
Patent Information
- Application Number
- CN202611082426.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-21
- Publication Date
- 2026-10-09
AI Technical Summary
本发明解决了现有核电厂阀门故障诊断过程高频信号干扰以及多源信息融合不足，故障诊断结果可解释性差的关键技术难题，以满足核电厂阀门持续健康监控的实际工程需求
[0052]本发明一种可解释性的核电厂阀门故障诊断方法、系统及介质，基于窄带包络解调算法和D-S证据理论进行阀门可解释性故障诊断。本发明方法通过使用时域分析和窄带包络解调算法对采集的高频传感数据进行处理，实现信号的解调和去噪，并计算特征参数，形成特征量集合。在此基础上，基于D-S证据理论实现多源传感数据的融合，进而根据融合结果实现阀门设备典型故障的决策级可解释性诊断。本发明解决了现有核电厂阀门故障诊断过程高频信号干扰以及多源信息融合不足，故障诊断结果可解释性差的关键技术难题，以满足核电厂阀门持续健康监控的实际工程需求。本发明成果能够为核电厂阀门设备的预测性维护提供参考依据，进而提升核电厂阀门的运行可靠性和智能化运维水平。对提高阀门等设备的运行安全性和经济性具有重要意义。
Smart Images
Figure CN122885461A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial equipment fault diagnosis technology, specifically to an interpretable method, system, and medium for diagnosing valve faults in nuclear power plants. Background Technology
[0002] Valves are the most frequently used and frequently operated general-purpose equipment in nuclear power plants. When valves malfunction, they often induce abnormal vibrations. Collecting and analyzing these abnormal vibrations is the first step in fault diagnosis technology. The raw vibration data contains rich fault information, but it also contains a large amount of interference unrelated to the fault. Therefore, signal processing techniques are needed to extract fault-related features. A major challenge in current signal processing technology is signal demodulation. Typical faults such as valve jamming, valve stem breakage, and seal leakage can cause signal modulation. Because the low-frequency impacts generated by the fault can excite high-frequency resonances in the equipment, the low-frequency frequencies directly related to the fault are submerged by the strong impact and are almost undetectable in the spectrum. Using appropriate signal demodulation techniques to extract the masked fault characteristic frequencies from the raw signal is a research hotspot in signal processing and feature extraction.
[0003] Meanwhile, for valve equipment fault diagnosis, focusing solely on vibration signals is often insufficient. While some fluid-induced faults (such as seal leaks) can cause abnormal vibrations, the fault characteristics of these vibration signals are not obvious and are often masked by the inherent vibrations of the mechanical equipment. In such cases, focusing on changes in parameters more sensitive to faults, such as flow rate, is more meaningful for valve equipment fault diagnosis. Furthermore, traditional fault diagnosis methods are usually based on signals collected by a single sensor. However, the information collected by a single sensor is limited, and fault feature extraction and identification based on this are often unreliable and cannot meet the needs of engineering applications. Therefore, multi-sensor technology has developed rapidly in recent years. Multi-sensor refers not only to the number of sensors (i.e., multiple sensors placed in different locations) but also to the types of sensors, such as using multiple sensors to simultaneously monitor vibration, temperature, and pressure signals. This provides richer information, but factors such as the equipment operating environment and human intervention can lead to uncertainties and even conflicts in multi-source information. Therefore, even with multi-sensor technology, how to synthesize all the information to make a diagnostic decision becomes a major challenge—that is, how to achieve multi-source information fusion from an algorithmic perspective.
[0004] Therefore, the existing valve fault diagnosis process in nuclear power plants suffers from high-frequency signal interference and insufficient fusion of multi-source information, resulting in a key technical challenge of poor interpretability of fault diagnosis results. Summary of the Invention
[0005] The technical problem this invention aims to solve is the key technical challenge of poor interpretability of fault diagnosis results due to high-frequency signal interference and insufficient multi-source information fusion in existing nuclear power plant valve fault diagnosis processes. The purpose of this invention is to provide an interpretable nuclear power plant valve fault diagnosis method, system, and medium, based on narrowband envelope demodulation algorithms and DS evidence theory for interpretable valve fault diagnosis. This invention's method processes acquired high-frequency sensor data using time-domain analysis and narrowband envelope demodulation algorithms to achieve signal demodulation and denoising, and calculates characteristic parameters to form a set of characteristic quantities. Based on this, DS evidence theory is used to fuse multi-source sensor data, and then the fusion results are used to achieve decision-level interpretable diagnosis of typical valve equipment faults. This invention solves the key technical problems of high-frequency signal interference and insufficient multi-source information fusion in existing nuclear power plant valve fault diagnosis processes, resulting in poor interpretability of fault diagnosis results, thus meeting the practical engineering needs of continuous health monitoring of nuclear power plant valves.
[0006] This invention is achieved through the following technical solution:
[0007] In a first aspect, the present invention provides an interpretable method for diagnosing valve faults in nuclear power plants, the method comprising:
[0008] The raw data of nuclear power plant valves is acquired and valve sample data is generated. The raw data includes vibration signals, differential pressure signals, current signals and flow signals. Among them, vibration signals are high-frequency signals, while differential pressure signals, current signals and flow signals are low-frequency signals.
[0009] The vibration signal is demodulated using a narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal; and time-domain features and frequency-domain features are extracted from the squared envelope spectrum to form a set of features.
[0010] Based on the feature set, differential pressure signal, current signal, and flow signal, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set, generating the BPA of all sensors; and the BPA of all sensors is fused based on DS evidence theory to obtain the final fault diagnosis result.
[0011] Furthermore, the method also includes: preprocessing the original data to obtain preprocessed original data.
[0012] Furthermore, the vibration signal is demodulated using a narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal, including:
[0013] The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands.
[0014] Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter.
[0015] The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal and use it as the real part signal.
[0016] The filtered vibration signal is subjected to Hilbert transform to obtain its suitable imaginary part signal;
[0017] The imaginary and real parts of the signal are constructed into an analytic signal, and the squared envelope spectrum of the signal is obtained based on the analytic signal.
[0018] Furthermore, the 1 / 2 binary tree and FIR filter bank include multiple levels of filters, with each level containing a low-pass filter and a high-pass filter.
[0019] No. Layer vibration signal After filtering by high-pass and low-pass filters at each stage, and then downsampling by 1 / 2, the result is obtained. Vibration signals of the layer and ,in, Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal at that location; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal at that location.
[0020] Furthermore, center frequency ,bandwidth The expression is:
[0021] ;
[0022] In the formula, The sampling frequency of the vibration signal. The value is the number of decomposition layers of the filter bank, and is a non-negative integer. ); i is the frequency band index of the filter bank, with a value range of . .
[0023] Furthermore, the kurtosis value is the fourth-order normalized cumulative amount of the subband complex envelope signal.
[0024] Furthermore, the signal squared envelope spectrum The expression is:
[0025] ;
[0026] In the formula, To analyze the signal; This is the filtered vibration signal, i.e., the real part of the signal; This is the imaginary part of the signal.
[0027] Furthermore, the time-domain features include waveform factor, impulse factor, skewness, and kurtosis;
[0028] Frequency domain characteristics include centroid frequency, root mean square frequency, and frequency standard deviation.
[0029] Furthermore, based on the feature set, differential pressure signal, current signal, and flow rate signal, Mahalanobis distance is used to evaluate and calculate the similarity between the training and test sets, generating the BPA for all sensors, including:
[0030] The valve sample data is randomly divided into a training set and a test set according to a preset ratio;
[0031] Based on the feature set, a reference feature matrix consisting of the feature set, differential pressure signal, current signal, and flow rate signal is constructed from the training set;
[0032] Simultaneously, a test vector consisting of a set of characteristic quantities, differential pressure signal, current signal, and flow rate signal is constructed from the test set;
[0033] Based on the reference feature matrix and the test vector, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set; after normalizing all the similarities of the sensors, the BPA corresponding to the sensor is obtained.
[0034] Following the above process, generate the BPA for all sensors.
[0035] Furthermore, based on DS evidence theory, the BPA of all sensors is fused to obtain the final fault diagnosis result, including:
[0036] Perform an orthogonal sum operation on the BPA of the first sensor and the BPA of the second sensor to obtain the first new BPA value;
[0037] The first new BPA value is orthogonally summed with the BPA of the third sensor to obtain the second new BPA value;
[0038] Following the above process, until the BPA calculation of the last sensor is completed, the final new BPA value is obtained;
[0039] The final fault diagnosis result is obtained by fusing all BPA values using the Dempster combination rule.
[0040] Secondly, the present invention provides an interpretable nuclear power plant valve fault diagnosis system, the system comprising:
[0041] The acquisition unit is used to acquire raw data of valves in nuclear power plants and form valve sample data. The raw data includes vibration signals, differential pressure signals, current signals and flow signals.
[0042] The demodulation and feature extraction unit is used to demodulate the vibration signal based on the narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal; and to extract time-domain features and frequency-domain features from the squared envelope spectrum of the signal to form a set of feature quantities.
[0043] The DS evidence theory fault diagnosis unit is used to calculate the similarity between the training set and the test set based on the feature set, differential pressure signal, current signal and flow signal using Mahalanobis distance, generate the BPA of all sensors, and fuse the BPA of all sensors based on DS evidence theory to obtain the final fault diagnosis result.
[0044] Furthermore, the demodulation processing and feature extraction unit includes a demodulation subunit, and the execution process of the demodulation subunit is as follows:
[0045] The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands.
[0046] Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter.
[0047] The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal and use it as the real part signal.
[0048] The filtered vibration signal is subjected to Hilbert transform to obtain its suitable imaginary part signal;
[0049] The imaginary and real parts of the signal are constructed into an analytic signal, and the squared envelope spectrum of the signal is obtained based on the analytic signal.
[0050] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described interpretable method for diagnosing valve faults in a nuclear power plant.
[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0052] This invention discloses an interpretable method, system, and medium for diagnosing valve faults in nuclear power plants. It utilizes a narrowband envelope demodulation algorithm and DS evidence theory for interpretable valve fault diagnosis. The method processes acquired high-frequency sensor data using time-domain analysis and a narrowband envelope demodulation algorithm to demodulate and denoise the signals, and calculates characteristic parameters to form a set of characteristic quantities. Based on this, DS evidence theory is used to fuse multi-source sensor data, and the fusion results enable decision-level interpretable diagnosis of typical valve equipment faults. This invention solves the key technical challenges of high-frequency signal interference and insufficient multi-source information fusion in existing nuclear power plant valve fault diagnosis processes, resulting in poor interpretability of fault diagnosis results, thus meeting the practical engineering needs of continuous health monitoring of nuclear power plant valves. The results of this invention can provide a reference for predictive maintenance of nuclear power plant valve equipment, thereby improving the operational reliability and intelligent operation and maintenance level of nuclear power plant valves. It is of great significance for improving the operational safety and economy of valves and other equipment.
[0053] (1) Improved accuracy of fault location: By integrating signal demodulation and multi-source data fusion diagnosis, the contribution weight of fault characteristics can be clearly displayed, enabling maintenance personnel to intuitively understand the diagnostic basis and reduce the misjudgment rate.
[0054] (2) Transparency of decision-making process: The diagnosis is based on multi-source data fusion based on DS evidence theory, which changes the black box diagnosis characteristics of traditional complex models such as neural networks and transforms the output of the diagnosis results into traceable rules, which meets the stringent requirements of nuclear power safety culture for operational transparency.
[0055] (3) Reduce operation and maintenance costs: shift from "periodic maintenance" to predictive maintenance to reduce over-maintenance costs and spare parts inventory costs. Attached Figure Description
[0056] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:
[0057] Figure 1 This is a flowchart illustrating an interpretable method for diagnosing valve faults in a nuclear power plant, as per the present invention.
[0058] Figure 2 This is a schematic diagram of the low-pass filter and high-pass filter structure for narrowband envelope demodulation according to the present invention;
[0059] Figure 3 This is a diagram of the filter bank structure for narrowband envelope demodulation according to the present invention;
[0060] Figure 4 This is a flowchart of the fault diagnosis process based on the DS evidence theory of this invention.
[0061] Figure 5This is a structural block diagram of an interpretable nuclear power plant valve fault diagnosis system according to the present invention. Detailed Implementation
[0062] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.
[0063] Existing nuclear power plant valve fault diagnosis processes suffer from high-frequency signal interference and insufficient multi-source information fusion, resulting in poor interpretability of fault diagnosis results. To address these challenges in signal processing and fault diagnosis, this invention proposes a multi-source heterogeneous fault characterization method based on narrowband envelope demodulation and DS evidence theory. Narrowband envelope demodulation is a signal processing technique used to study the evolution of signals in time and frequency, and is particularly suitable for signal demodulation. DS evidence theory is one of the most representative random set theories; essentially, it is an uncertainty reasoning method for decision fusion. It not only effectively expresses the uncertainty of information but also completes information fusion without prior information.
[0064] Therefore, based on the above, this invention designs an interpretable method, system, and medium for diagnosing valve faults in nuclear power plants. It utilizes a narrowband envelope demodulation algorithm and DS evidence theory for interpretable valve fault diagnosis. The method processes acquired high-frequency sensor data using time-domain analysis and a narrowband envelope demodulation algorithm to demodulate and denoise the signal, and calculates characteristic parameters to form a set of feature quantities. Based on this, DS evidence theory is used to fuse multi-source sensor data, and the fusion results enable decision-level interpretable diagnosis of typical valve equipment faults. This invention solves the key technical problems of high-frequency signal interference and insufficient multi-source information fusion in existing nuclear power plant valve fault diagnosis processes, resulting in poor interpretability of fault diagnosis results, thus meeting the practical engineering needs of continuous health monitoring of nuclear power plant valves. The results of this invention can provide a reference for predictive maintenance of nuclear power plant valve equipment, thereby improving the operational reliability and intelligent operation and maintenance level of nuclear power plant valves.
[0065] Example 1
[0066] like Figure 1 As shown, the present invention provides an interpretable method for diagnosing valve faults in nuclear power plants, the method comprising:
[0067] Step 1: Obtain raw data of nuclear power plant valves and form valve sample data. Raw data includes vibration signals, differential pressure signals, current signals and flow signals.
[0068] In this embodiment, vibration, pressure, current, and flow signals of the nuclear power plant valves are acquired by sensors installed at the valves and stored in a computer. The vibration signal is a high-frequency signal, while the pressure difference signal, current signal, and flow signal are low-frequency signals.
[0069] In this embodiment, the method further includes: preprocessing the original data to obtain preprocessed original data. Preprocessing includes outlier removal and signal mean removal.
[0070] In the above technical solution, step 1 of the present invention is data acquisition and processing: the raw data obtained from process parameter sensors such as vibration signals, differential pressure signals, current signals and flow signals on the valves of the nuclear power plant are stored in the computer through a data acquisition board, and the acquired raw data is preprocessed at the same time.
[0071] Step 2: Demodulate the preprocessed vibration signal based on the narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal; and extract time-domain features and frequency-domain features from the squared envelope spectrum of the signal to form a set of feature quantities.
[0072] In this embodiment, to address the problem that high-frequency fault signals from valve vibration in nuclear power plants are overwhelmed by noise and cannot be extracted, demodulation processing of the vibration signals is performed. Step 2 specifically includes:
[0073] (1) The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands;
[0074] Specifically, the 1 / 2 binary tree and FIR filter bank include multiple stages of filters, such as Figure 2 As shown, Figure 2 A schematic diagram of the low-pass and high-pass filter structures for narrowband envelope demodulation; each stage of the single filter includes a low-pass filter h0(n) and a high-pass filter h1(n), represented as:
[0075] ;
[0076] In the formula: h(n) is the index of the discrete-time series, and h(n) is the mother filter sequence.
[0077] like Figure 3 As shown, Figure 3 This is a diagram of a filter bank structure for narrowband envelope demodulation, consisting of several individual filters forming the filter bank. Layer vibration signal After filtering by high-pass and low-pass filters at each stage, and then downsampling by 1 / 2, the result is obtained. Vibration signals of the layer and ,in, Vibration signal In the The layer is located at the center frequency ,bandwidth The complex envelope of the time-domain signal at that location; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal at that location.
[0078] Specifically, center frequency ,bandwidth The two are represented as follows:
[0079] ;
[0080] In the formula, The sampling frequency of the vibration signal. The value is the number of decomposition layers of the filter bank, and is a non-negative integer. ); i is the frequency band index of the filter bank, with a value range of . .
[0081] (2) Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter.
[0082] Specifically, the kurtosis value is the sub-band complex envelope signal. The fourth-order normalized cumulative quantity, kurtosis value The expression is:
[0083] ;
[0084] In the formula, the constant -2 at the end is to make the complex envelope of the Gaussian signal zero.
[0085] Specifically, when the pulse fault frequency component dominates in the signal, the kurtosis value is often very high. Therefore, spectral kurtosis can be used as an evaluation index to find the resonant frequency band containing the richest fault impact information. Thus, the frequency band with the largest kurtosis value is selected, and its center is... ,bandwidth The center frequency band and bandwidth of the optimal filter.
[0086] (3) The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal as the real part signal; the Hilbert transform is performed on the filtered vibration signal to obtain its suitable imaginary part signal; the imaginary part signal The expression is:
[0087] ;
[0088] In the formula, This is the filtered vibration signal, i.e., the real part of the signal.
[0089] (4) Construct the imaginary part and the real part of the signal into an analytic signal, and obtain the squared envelope spectrum of the signal based on the analytic signal.
[0090] Specifically, the imaginary part signal and real part signal Construct as an analytic signal , is represented as:
[0091] ;
[0092] Envelope signal is the analytic signal The modulus, and the squared envelope is the square of the envelope, the signal squared envelope spectrum. The expression is:
[0093] ;
[0094] In the formula, To analyze the signal; This is the filtered vibration signal, i.e., the real part of the signal; This is the imaginary part of the signal.
[0095] (5) Based on the signal square envelope spectrum ,from Four time-domain features (waveform factor, impact factor, skewness, and kurtosis) and three frequency-domain features (centroid frequency, root mean square frequency, and frequency standard deviation) are extracted from the vibration signal to form the vibration signal. The set of features.
[0096] In the above technical solution, step 2 of this invention involves signal processing and feature extraction: Based on a narrowband envelope demodulation algorithm, the signal modulation mechanism of valve fault components is analyzed, a monitoring signal modulation process is established, and signal demodulation and denoising are achieved. On this basis, according to the acquired signal square envelope spectrum, four time-domain features (waveform factor, impulse factor, skewness, and kurtosis) and three frequency-domain features (centroid frequency, root mean square frequency, and frequency standard deviation) of the multi-source high-frequency signal are extracted to characterize valve fault information. This invention extracts frequency-domain features such as centroid frequency and root mean square frequency from the demodulated signal square envelope spectrum, enhancing the correlation between frequency-domain features and faults, and solving the problem of fault signals being submerged and unable to be extracted.
[0097] Step 3: Based on the feature set, differential pressure signal, current signal, and flow signal, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set to generate the BPA of all sensors; and the BPA of all sensors is fused based on DS evidence theory to obtain the final fault diagnosis result.
[0098] BPA stands for Basic Probability Assignment.
[0099] In this embodiment, based on the feature set, differential pressure signal, current signal, and flow rate signal, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set, generating the BPA for all sensors, including:
[0100] Step 31: Randomly divide the valve sample data (preprocessed vibration signal, differential pressure signal, current signal, and flow signal) into training set and test set according to a preset ratio;
[0101] Step 32: Based on the feature set, construct a reference feature matrix from the training set, which includes the feature set, differential pressure signal, current signal, and flow rate signal. , is represented as:
[0102] ;
[0103] In the formula, The number of valves representing the characteristics of a nuclear power plant. This represents the number of valve failure types in a nuclear power plant. For example, This represents the first characteristic of a Class I fault in a nuclear power plant valve.
[0104] At the same time, a test set is built using the same method. test vectors , is represented as:
[0105] ;
[0106] In the formula, These represent the features of the test vector.
[0107] Step 33: Based on the reference feature matrix and the test vector, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set; and after normalizing all the similarities of the sensors, the BPA corresponding to the sensor is obtained.
[0108] Specifically, Mahalanobis distance, by introducing the inverse of the covariance matrix and performing a linear transformation on the data, can effectively eliminate correlations and dimensional differences between features, making the distance metric more accurate in the standardized space. It is particularly robust and statistically significant compared to Euclidean distance when dealing with non-uniformly distributed or correlated features. Therefore, this invention uses Mahalanobis distance to evaluate the similarity between the training and test sets, generating the BPA of the evidence body (i.e., the BPA corresponding to the sensor). reference vectors and test vectors of the test set Mahalanobis distance between Represented as:
[0109] ;
[0110] In the formula, Reference feature matrix A vector composed of the data in the nth column.
[0111] Calculate the first Similarity between each reference vector and the test vector Based on the distance between the two, we can calculate:
[0112] ;
[0113] Suppose we are currently dealing with the first... Data from each sensor, including all similarities of that sensor. ( Normalize to Then, the BPA corresponding to the sensor can be obtained:
[0114] ;
[0115] In the formula, ( ) is the first The BPA corresponding to each sensor.
[0116] Step 34: Following the above process, generate the BPA for all sensors.
[0117] In this embodiment, the BPA of all sensors is fused based on DS evidence theory to obtain the final fault diagnosis result, including:
[0118] For multiple pieces of evidence (sensors), the first step is to perform an orthogonal sum operation on the BPA of the first and second sensors to obtain a first new BPA value, which is the result of combining the first two pieces of evidence. Then, the first new BPA value is orthogonally summed with the BPA of the third sensor (i.e., the third piece of evidence) to obtain a second new BPA value. This process continues until the BPA of the last sensor is calculated, resulting in the final new BPA value, which is the result of combining all the evidence. Finally, all BPA values are fused using the Dempster combination rule to obtain the fusion result m(F) that integrates all information sources.
[0119] if The following conditions must be met:
[0120] ;
[0121] Then F1 is the final fault diagnosis result.
[0122] In the above technical solution, step 3 of this invention is an interpretable fault diagnosis based on DS evidence theory: Based on the feature set of signal processing and feature extraction in step 2, the probabilistic correspondence between fault mode (identification framework) and fault symptom (modulation intensity index) is explored from the perspective of probability statistics and set theory. Under the system of DS evidence theory, the fusion diagnosis of multiple sensors is completed by calculating the similarity of feature vectors of training set and test set.
[0123] This invention, based on the DS evidence theory, generates a BPA (Browser Apparent Differential) using Mahalanobis distance, eliminating the correlation between various characteristics and making them independent, thus enhancing the interpretation of small changes in variables. Interpretable fault diagnosis based on DS evidence theory can provide a probabilistic expression of the diagnostic results, while also providing the diagnostic contribution of each sensor, thereby improving the interpretability of the fault diagnosis process.
[0124] Example 2
[0125] like Figure 5 As shown, Figure 5 This is a structural block diagram of an interpretable nuclear power plant valve fault diagnosis system according to the present invention. The difference between this embodiment and Embodiment 1 is that this embodiment provides an interpretable nuclear power plant valve fault diagnosis system, which corresponds one-to-one with the interpretable nuclear power plant valve fault diagnosis method of Embodiment 1. The system includes:
[0126] The acquisition unit is used to acquire raw data of valves in nuclear power plants and form valve sample data. The raw data includes vibration signals, differential pressure signals, current signals and flow signals.
[0127] The demodulation and feature extraction unit is used to demodulate the vibration signal based on the narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal; and to extract time-domain features and frequency-domain features from the squared envelope spectrum of the signal to form a set of feature quantities.
[0128] The DS evidence theory fault diagnosis unit is used to calculate the similarity between the training set and the test set based on the feature set, differential pressure signal, current signal and flow signal using Mahalanobis distance, generate the BPA of all sensors, and fuse the BPA of all sensors based on DS evidence theory to obtain the final fault diagnosis result.
[0129] As a further implementation, the demodulation processing and feature extraction unit includes a demodulation subunit, the execution process of which is as follows:
[0130] The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands.
[0131] Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter.
[0132] The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal and use it as the real part signal.
[0133] The filtered vibration signal is subjected to Hilbert transform to obtain its suitable imaginary part signal;
[0134] The imaginary and real parts of the signal are constructed into an analytic signal, and the squared envelope spectrum of the signal is obtained based on the analytic signal.
[0135] The execution process of each unit can be carried out according to the steps of an interpretable nuclear power plant valve fault diagnosis method in Example 1, and will not be described in detail in this example.
[0136] Meanwhile, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described interpretable method for diagnosing valve faults in nuclear power plants.
[0137] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0138] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0141] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. An interpretable method for diagnosing valve faults in nuclear power plants, characterized in that, The method includes: Acquire raw data of nuclear power plant valves and form valve sample data, wherein the raw data includes vibration signals, differential pressure signals, current signals and flow signals; The vibration signal is demodulated using a narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal; and time-domain features and frequency-domain features are extracted from the squared envelope spectrum of the signal to form a set of features. Based on the set of features, differential pressure signal, current signal, and flow signal, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set, generating the BPA of all sensors; and the BPA of all sensors is fused based on DS evidence theory to obtain the final fault diagnosis result.
2. The interpretable nuclear power plant valve fault diagnosis method according to claim 1, characterized in that, The method further includes: preprocessing the original data to obtain preprocessed original data.
3. The interpretable nuclear power plant valve fault diagnosis method according to claim 1, characterized in that, The vibration signal is demodulated using a narrowband envelope demodulation algorithm to obtain the squared envelope spectrum of the signal, including: The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands. Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter. The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal, which is then used as the real part signal. Perform a Hilbert transform on the filtered vibration signal to obtain the imaginary part of the signal; The imaginary part signal and the real part signal are constructed into an analytic signal, and the squared envelope spectrum of the signal is obtained based on the analytic signal.
4. The interpretable nuclear power plant valve fault diagnosis method according to claim 3, characterized in that, The 1 / 2 binary tree and FIR filter bank include multiple levels of filters, and each level of filter includes a low-pass filter and a high-pass filter. No. Layer vibration signal After filtering by high-pass and low-pass filters at each stage, and then downsampling by 1 / 2, the result is obtained. Vibration signals of the layer and ,in, Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal at that location; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal; Vibration signal In the The layer is located at the center frequency The complex envelope of the time-domain signal at that location.
5. The interpretable nuclear power plant valve fault diagnosis method according to claim 4, characterized in that, The center frequency ,bandwidth The expression is: ； In the formula, The sampling frequency of the vibration signal. This represents the number of decomposition layers in the filter bank, and its value is a non-negative integer. ; i is the frequency band index of the filter bank, and its value ranges from 1 to 2. .
6. The interpretable nuclear power plant valve fault diagnosis method according to claim 3, characterized in that, The kurtosis value is the fourth-order normalized cumulative value of the subband complex envelope signal.
7. The interpretable nuclear power plant valve fault diagnosis method according to claim 3, characterized in that, The signal square envelope spectrum The expression is: ； In the formula, To analyze the signal; This is the filtered vibration signal, i.e., the real part of the signal; This is the imaginary part of the signal.
8. The interpretable nuclear power plant valve fault diagnosis method according to claim 1, characterized in that, The time-domain features include waveform factor, impulse factor, skewness, and kurtosis; The frequency domain features include the centroid frequency, root mean square frequency, and frequency standard deviation.
9. The interpretable method for diagnosing valve faults in a nuclear power plant according to claim 1, characterized in that, The step involves using Mahalanobis distance to evaluate and calculate the similarity between the training and test sets based on the set of features, differential pressure signal, current signal, and flow signal, and generating a BPA for all sensors, including: The valve sample data is randomly divided into a training set and a test set according to a preset ratio; Based on the set of features, a reference feature matrix is constructed from the training set, comprising the set of features, differential pressure signal, current signal, and flow rate signal. Simultaneously, a test vector consisting of the set of characteristic quantities, differential pressure signal, current signal, and flow rate signal is constructed from the test set; Based on the reference feature matrix and the test vector, Mahalanobis distance is used to evaluate and calculate the similarity between the training set and the test set; and after normalizing all the similarities of the sensors, the BPA corresponding to the sensor is obtained. Following the above process, generate the BPA for all sensors.
10. The interpretable method for diagnosing valve faults in a nuclear power plant according to claim 1, characterized in that, Based on the DS evidence theory, the BPA of all sensors is fused to obtain the final fault diagnosis results, including: Perform an orthogonal sum operation on the BPA of the first sensor and the BPA of the second sensor to obtain the first new BPA value; The first new BPA value is orthogonally summed with the BPA of the third sensor to obtain the second new BPA value; Following the above process, until the BPA calculation of the last sensor is completed, the final new BPA value is obtained; The final fault diagnosis result is obtained by fusing all BPA values using the Dempster combination rule.
11. An interpretable nuclear power plant valve fault diagnosis system, characterized in that, The system includes: The acquisition unit is used to acquire raw data of valves in nuclear power plants and form valve sample data. The raw data includes vibration signals, differential pressure signals, current signals and flow signals. The demodulation processing and feature extraction unit is used to demodulate the vibration signal based on the narrowband envelope demodulation algorithm to obtain the signal squared envelope spectrum; and to extract time-domain features and frequency-domain features from the signal squared envelope spectrum to form a feature set. The DS evidence theory fault diagnosis unit is used to calculate the similarity between the training set and the test set based on the set of features, differential pressure signal, current signal and flow signal, using Mahalanobis distance to generate the BPA of all sensors; and to fuse the BPA of all sensors based on DS evidence theory to obtain the final fault diagnosis result.
12. The interpretable nuclear power plant valve fault diagnosis system according to claim 11, characterized in that, The demodulation processing and feature extraction unit includes a demodulation subunit, and the execution process of the demodulation subunit is as follows: The vibration signal is filtered and decomposed based on a 1 / 2 binary tree and an FIR filter bank to obtain sub-band complex envelope signals of several frequency bands. Calculate the kurtosis value of the sub-band complex envelope signal of each frequency band and select the frequency band with the largest kurtosis value. Then, take the center and bandwidth of the selected frequency band as the center frequency band and bandwidth of the optimal filter. The vibration signal is filtered based on the optimal filter to obtain the filtered vibration signal, which is then used as the real part signal. Perform a Hilbert transform on the filtered vibration signal to obtain the imaginary part of the signal; The imaginary part signal and the real part signal are constructed into an analytic signal, and the squared envelope spectrum of the signal is obtained based on the analytic signal.
13. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements an interpretable method for diagnosing valve faults in a nuclear power plant as described in any one of claims 1 to 10.